add custom handler

handler.py

@@ -0,0 +1,29 @@
+from typing import Dict, List, Any
+from transformers import CLIPModel, AutoProcessor, AutoTokenizer
+import torch
+from PIL import Image
+import requests
+
+
+class EndpointHandler:
+    def __init__(self):
+        self.model = CLIPModel.from_pretrained("patrickjohncyh/fashion-clip")
+        self.processor = AutoProcessor.from_pretrained("patrickjohncyh/fashion-clip")
+        self.tokenizer = AutoTokenizer.from_pretrained("openai/clip-vit-base-patch32")
+
+    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
+        parameters = data.pop("parameters", {"mode": "image"})
+        inputs = data.pop("inputs", data)
+        with torch.no_grad():
+            if parameters["mode"] == "text":
+                inputs = self.tokenizer(inputs, padding=True, return_tensors="pt")
+                features = self.model.get_text_features(**inputs)
+
+            if parameters["mode"] == "image":
+                url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+                image = Image.open(requests.get(url, stream=True).raw)
+
+                inputs = self.processor(images=image, return_tensors="pt")
+                features = self.model.get_image_features(**inputs)
+
+            return features[0].tolist()

requirements.txt

@@ -0,0 +1 @@
+Pillow

test_handler.py

@@ -0,0 +1,22 @@
+from handler import EndpointHandler
+
+# init handler
+my_handler = EndpointHandler()
+
+# prepare sample payload
+non_holiday_payload = {
+    "inputs": "I am quite excited how this will turn out",
+    "parameters": {"mode": "text"},
+}
+
+# test the handler
+non_holiday_pred = my_handler(non_holiday_payload)
+print(non_holiday_pred)
+
+non_holiday_payload = {
+    "inputs": "https://image.momoxfashion.com/Marc-O-Polo-i402d2w-0-detail",
+    "parameters": {"mode": "image"},
+}
+
+non_holiday_pred = my_handler(non_holiday_payload)
+print(non_holiday_pred)